OpenAI's defence rests on 14 instances

On 4 September both sides in the consolidated OpenAI copyright litigation asked a federal judge to decide the case without a trial. Read the two briefs together and they barely touch.

The authors' memorandum opens by saying OpenAI "built the foundations of its business on mass piracy" — and adds that the defendants "do not dispute" much of what follows. Their case is about acquisition: how the books entered the training data in the first place.

OpenAI's memorandum is about what comes out the other end. It argues the use was transformative and that the plaintiffs' own evidence amounts to "an alleged regurgitation rate of 0.00007%".

Open the filing and that number has a denominator problem. The same sentence continues: "Of these 14 instances, all but…". The rate is not a measurement of how the model behaves in the wild. It is fourteen examples, divided by a very large number. OpenAI's Rule 56.1 statement attributes the figure to the declaration of its expert, Dr Taylor Berg-Kirkpatrick. Publishers Weekly additionally reports that the plaintiffs' expert, working at it, could extract a 1,899-word passage from A Game of Thrones — about 0.62% of the novel.

Those figures all measure output. They say nothing about acquisition. An argument that a machine rarely reproduces a book is not an answer to the claim that the book was pirated to build the machine; it is an answer to a different question. Whether Judge Sidney Stein treats those as one question or two is, more or less, the case.

He has already shown he will look at outputs closely. In October 2025 he held that ChatGPT-generated plot summaries of the plaintiffs' novels could themselves constitute infringement, refusing OpenAI summary judgment on the output claims. A court willing to find infringement in a summary is not obviously a court that will be moved by a decimal place.

There is a second fight running underneath the first. Much of the supporting record — declaration after declaration of exhibits — is filed under seal. On 21 September, Bloomberg Industry Group moved to intervene for the limited purpose of opposing that sealing, and Magistrate Judge Ona Wang ordered the parties to propose a briefing schedule by 25 September. How much of this evidence ever becomes public is, right now, itself being litigated. Separately, Microsoft has asked for more time to oppose the news plaintiffs' motion, so the schedule is already moving.

The thirteen named authors are not a random draw: David Baldacci, Taylor Branch, Michael Connelly, Sylvia Day, Jonathan Franzen, Christopher Golden, Andrew Sean Greer, John Grisham, David Henry Hwang, George R.R. Martin, Jodi Picoult, Stacy Schiff and James Shapiro. Commercial fiction is carrying this, not the literary end, and Sylvia Day's name is the one romance should notice.

Responses are due 9 October.

Filings cited: ECF 1982 and ECF 1983, In Re: OpenAI, Inc. Copyright Infringement Litigation, 1:25-md-03143 (S.D.N.Y.).